Release 5 of Petalpelago is now available!

Penny finds an interesting computer game to play, gets a chance to talk with Mina, and makes some surprising revelations when she hears noises coming from the reading lounge...

This release brings the total word count of the story to just shy of 38K words.

Enjoy!

~

What follows are some personal thoughts and a little behind-the-scenes info about this release for those who are curious...

I tried a different approach to images with this release, with two different ways that I went about layering them. The main difference is instead of trying to generate character(s) plus a background image all in a single prompt to the AI, which very often failed to meet my goals for the image and scene for a variety of reasons, I tried generating scenery images separately and adding individually generated character images to them manually myself.

I had done something similar with the earlier image for Penny on the toilet, combining multiple images together, removing backgrounds, and editing out certain parts of them. It's a lot more manual time and care needed in order to get the image looking like I had envisioned, but just relying solely on AI prompts or even the img2img feature with in-painting sometimes still didn't get the results I had in mind. At times the AI just doesn't quite understand how to show an image of what you've got in mind.

The two types of approaches to layering that I used for this update are in Penny's image and the scene in the reading lounge. One has the "glowing" white border around it, which is the same approach that I tend to take for the Petals title screen images, and the other was a more manual, direct way of combining images together.

The "glowing" one looks nicer in that there are no jagged edges left around it, but of course it has that glowing look to it which separates it from the background a bit, making it feel a little more artificial than I'd prefer. The second way takes a lot more time to do, but it puts characters more directly into the scene without the glowing visuals around them.

Now, I'll be the first to admit my image editing skills are novice/intermediate level at best. I'm not sure which approach I like the results of more. Neither is perfect, and I think both have their pluses and minuses. Overall though, I am feeling more satisfied with the looks of the characters themselves by having the AI generate each character on their own separately, and then adding them to the scenery image manually.

I also tried putting the manually-edited images back into the AI and getting it to generate variations, but the results were mixed. The image with the glowing effect fared much better, but that might just be because the AI struggles overall when you try to generate images with multiple characters in it.

It may just be a learning curve for me too, with gradually understanding how to use the AI more effectively over time, as I'm still pretty new to all this. Or just improving my image editing skills as I go along, to help the scenes look a little more natural as less amateurish.

I think one way or another I will continue in this fashion for the "realistic portrayal" images that I struggle to get good results from the AI directly. For the scenes that take place in the mysterious petal realm, my current thought is to continue to generate those in a single prompt to the AI. They are very detailed and intricate by comparison, and I think it would lose some of the visual appeal if I tried either of these layering approaches with those images.